Oilers analytic/advanced stat thread

[Stoneman89]

Corsi is only looked at when at 5v5, basically.

This team isn't losing because of 5v5. It's special teams. They've got a bottom 5 PP and a historically bad PK.

To put the PK in perspective, the Oilers are scored on just over 4.5 times the rate that LA's(#1 PK) is scored on.

Guess Corsi isn't the end all to be able to judge a teams performance. As we all saw last year, the Kings either led the league or were near the top in Corse and finished out of the playoffs.

 

[McDNicks17]

Guess Corsi isn't the end all to be able to judge a teams performance. As we all saw last year, the Kings either led the league or were near the top in Corse and finished out of the playoffs.

They were in the same boat as the Oilers are now.

High CF%, league worst SH% and middle of the pack SV% at 5v5 along with a terrible PK.

 

[Stoneman89]

They were in the same boat as the Oilers are now.

High CF%, league worst SH% and middle of the pack SV% at 5v5 along with a terrible PK.

There you go.

 

[McDNicks17]

I think a big reason for the Oilers high CF% is because of how much time they've spent trailing.

Teams tend to have a big push in CF% when trailing and the Oilers have spent almost twice as much time at 5v5 trailing(258:48) as leading(131:25).

Oilers Trailing: 60.07 CF%
Oilers Leading: 49.02 CF%


Some other examples:

Carolina Trailing: 66.33 CF% (1st in the league)
Carolina Leading: 46.78 CF%

Minnesota Trailing: 61.46 CF% (2nd in the league)
Minnesota Leading: 38.95 CF%

 

[CantHaveTkachev]

I think a big reason for the Oilers high CF% is because of how much time they've spent trailing.

Teams tend to have a big push in CF% when trailing and the Oilers have spent almost twice as much time at 5v5 trailing(258:48) as leading(131:25).

Oilers Trailing: 60.07 CF%
Oilers Leading: 49.02 CF%


Some other examples:

Carolina Trailing: 66.33 CF% (1st in the league)
Carolina Leading: 46.78 CF%

Minnesota Trailing: 61.46 CF% (2nd in the league)
Minnesota Leading: 38.95 CF%

who's got the best corsi when leading?

 

[CycloneSweep]

who's got the best corsi when leading?

Only 2 teams have an over 50% corsi when leading. Montreal and Calgary. Oilers are 5th

 

[CycloneSweep]

We have the second best fenwick when leading and the 7th best when trailing.

 

[Stoneman89]

I think a big reason for the Oilers high CF% is because of how much time they've spent trailing.

Teams tend to have a big push in CF% when trailing and the Oilers have spent almost twice as much time at 5v5 trailing(258:48) as leading(131:25).

Oilers Trailing: 60.07 CF%
Oilers Leading: 49.02 CF%


Some other examples:

Carolina Trailing: 66.33 CF% (1st in the league)
Carolina Leading: 46.78 CF%

Minnesota Trailing: 61.46 CF% (2nd in the league)
Minnesota Leading: 38.95 CF%

Losing teams tend to trail a lot.. Makes sense.

 

[Stoneman89]

We have the second best fenwick when leading and the 7th best when trailing.

Someone should let TM know about this. I'm sure it will warm his heart and give him some comfort.

 

[CycloneSweep]

Someone should let TM know about this. I'm sure it will warm his heart and give him some comfort.

Something something goals will come.

 

[MessierII]

Except even old Corsi has proven more predictive than raw goals since goals are basically random events.

Here’s a very detailed article as to why this statement is 100% wrong. Goal based metrics are better at everything.

Goal-based Metrics Better Than Shot-based Metrics at Predicting Hockey Success

 

[nexttothemoon]

If you use some logic you can see that goals for and against are what actually counts on the scoreboard and is actually what wins and loses games.

Corsi, Fenwick, Shots and any other derivative stats are just tangential to what actually happens on the scoreboard and are going to be less useful when you don't take into account GF and GA. Certainly it's nice to have a pile of shots... but not if you can't score on enough of those shots. Same goes for having a low number of shots against as well... which is also useless if you are letting in a high percentage of those shots as well.

This is all pretty elementary stuff and it really isn't as complicated as most make it out to be. This isn't all so hard to figure out that it makes everyone throw their hands in the air and say it's an enigma wrapped up in a mystery.

You score more than your opponent... you win. You score less than your opponent... you lose.

You have players that are on the ice for more goals for than goals against... you'll win more games... you have players that are on the ice for more goals against than goals for... you'll lose more games.

PDO isn't magic or "luck". It's simply shooting percentage + save percentage. A team (or group of players) with low PDO will lose more games than they will win unless they have many more shots for than shots against to overcome that combo of low shooting percentage and low save percentage.

Teams with high positive corsi/fenwick/shots differentials miss the playoffs far more often than teams with high PDO and high GF/GA differentials.

Right now among the bottom 12 teams in PDO in the NHL...

Worst 12 PDO teams:

Buffalo
Arizona
Edmonton
Carolina
Montreal
Pittsburgh x
Philadelphia
Detroit
Florida
Ottawa
Calgary x
Boston x

Is it just unlucky that 9 of those teams also are among the bottom 12 teams in the league in points% so far?

Pittsburgh, Boston and Calgary are the only exceptions currently.


Best 12 PDO teams:

Tampa
Winnipeg
Toronto
LA
NYI
Nashville
NYR x
Minnesota x
New Jersey
Chicago x
Vegas
Anaheim x


Only NYR, Minnesota, Chicago and Anaheim aren't in the top 12 teams in points % this season.

The correlation between wins and PDO is high and GF/GA differential to wins is even higher.


If that correlation exists on a team level... why do you think that magically disappears on an individual level.

Individual players obviously have different levels of ozone to dzone starts, quality of teammates/opposition, icetime and individual offensive contributions as well. That's why these variables have to be taken into account as well because on an individual level there is a correlation between those factors and overall individual GF/GA differential as well. So take those into account and then look at each players adjusted relative GF/GA ratio in comparison to their teammates.

That's all there is. It's not rocket science and it's not mysterious or impossible to fathom. It's not even advanced stats... it's just reasonable, logical evaluations of what actually is happening on the ice.

Advanced stats are meaningless if they don't keep in mind the reality of GF/GA on the scoreboard as that's what wins and loses game. The farther you deviate away from that... the more useless and tangential the stats become and the less they will reflect reality.

 

[Laineux]

Advanced stats are meaningless if they don't keep in mind the reality of GF/GA on the scoreboard as that's what wins and loses game. The farther you deviate away from that... the more useless and tangential the stats become and the less they will reflect reality.

The point of advanced stats is that they predict future GF/GA better than GF/GA alone.

 

[Little Fury]

Here’s a very detailed article as to why this statement is 100% wrong. Goal based metrics are better at everything.

Goal-based Metrics Better Than Shot-based Metrics at Predicting Hockey Success

There's some seriously flawed assumptions and half-baked methodology in that piece that's been discussed elsewhere. There's a much larger body of work to the contrary. I'm just gonna leave it at that.

 

[McDNicks17]

I think the idea that GF/GA and PDO have predictive value is just logically false.

Of course you're going to be a good team if everyone is shooting over league average and you're goalie is making saves above league average. If you have a team full of 2012 Eberles shooting 19%, you're probably going to win the Cup if your goalie isn't letting in every other shot.

SH% and SV% are sporadic. It's the reason why PDO is a stat in the first place. No player or team is going to have the same SH%/SV% from year to year and that's why it has basically no predictive value.

 

[nexttothemoon]

I look at it more so as stats that show reality. Any stats having predictive value are obviously not going to be 100% correct... because of the uncertainty of future slumps, injuries, hot streaks, roster changes, coaching changes etc. That's obviously outside the abilities of any stats to be able to show because the events of the future aren't here and haven't happened yet.

A team with high corsi/fenwick/shot differentials... or some other tangential stat (chances etc)... can stay high for an extended stretch and that team can still miss the playoffs because they are simply outscored by their opposition... due to many other factors.

Put it this way... anyone is welcome to come up with some current stat NOT based on GF/GA that will predict which 10 teams will FOR SURE make the playoffs this season.

I'll use the current highest/lowest GF/GA differentials calculated 3 different ways in the charts below:

Based on a dead simple stat of GF/GA ratio... those 10 teams are the best in the NHL right now...

Best 10 teams currently:

Tampa	1.424
LA	1.361
Winnipeg	1.31
St Louis	1.194
Toronto	1.181
Columbus	1.169
NYI	1.148
San Jose	1.143
Chicago	1.139
NYR	1.117

[TBODY] [/TBODY]

Based on a slightly modified GF/GA stat (SRS which uses GF/GA but also takes into account the strength of schedule for each team)...
2017-18 NHL Summary | Hockey-Reference.com

Best 10 teams currently:

Tampa Bay Lightning	1.02
Los Angeles Kings	0.88
Winnipeg Jets	0.82
St. Louis Blues	0.68
Columbus Blue Jackets	0.54
Toronto Maple Leafs	0.5
New York Islanders	0.49
Chicago Blackhawks	0.39
Nashville Predators	0.38
New York Rangers	0.37

[TBODY] [/TBODY]

Based on a combo of 3 different methods of looking at GF/GA ratios (but still completely score based) these are the current Sagarin RATINGS...
https://www.usatoday.com/sports/nhl/sagarin/

Best 10 teams currently:

Tampa	4.49
LA	4.34
Winnipeg	4.31
Toronto	4.27
Chicago	4.25
Columbus	4.24
Nashville	4.24
Dallas	4.22
St Louis	4.21
NYR	4.18

[TBODY] [/TBODY]

I contend that those lists (likely the 2nd or 3rd are best as they have slightly more combined data yet are still entirely based on GF/GA) are as good as it gets when it comes to "predicting" the future.
Those 10 teams on those lists should all make the playoffs... with the obvious proviso that dramatic injuries/slumps/roster changes could occur that alters the team drastically.

 

[MessierII]

There's some seriously flawed assumptions and half-baked methodology in that piece that's been discussed elsewhere. There's a much larger body of work to the contrary. I'm just gonna leave it at that.

There’s way more flawed methodology and half baked methodology on any analytics blog ill guarantee you that. Shot attempts have been absolute garbage at predicting anything for a few years running now.

 

[Little Fury]

There’s way more flawed methodology and half baked methodology on any analytics blog ill guarantee you that. Shot attempts have been absolute garbage at predicting anything for a few years running now.

 

[Vanqu1sh]

Analytics, by their nature, are not predictive. All analytics are pulled from on-ice events from each player, and thus, any future stats are created by what happens on the ice. If a player is playing 1st line minutes and has positive possession metrics, people will assume that they will continue to have these types of results. That's about as predictive as you watching a player play on the 1st line and see that they are talented enough to have success in that position.

The next year the player could be struggling because of some random issue, who knows what, now they dont look like a 1st line player, and their stats will represent that. Players success on the ice determines what their analytics look like, not the other way around.

 

[PerformanceMcOil]

I don't see Kris Russell on this chart anywhere...

 

[MessierII]

Not sure where those numbers are from they’re completely wrong.

Team Advanced Stats Finder | Hockey-Reference.com

LA had the best corsi for last year and didn’t even make the playoffs. In fact 5 of the top 10 corsi teams didn’t make the playoffs. The Stanley Cup finals featured the 8th best corsi for team vs the 15th best and the 15th best won. Last year was an unmitigated disaster for corsi as a predictive stat.

 

[Little Fury]

Not sure where those numbers are from they’re completely wrong.

Team Advanced Stats Finder | Hockey-Reference.com

LA had the best corsi for last year and didn’t even make the playoffs. In fact 5 of the top 10 corsi teams didn’t make the playoffs. The Stanley Cup finals featured the 8th best corsi for team vs the 15th best and the 15th best won. Last year was an unmitigated disaster for corsi as a predictive stat.

Chart was from here. Numbers aren't wrong, you're just not reading the chart right.

 

[MessierII]

Chart was from here. Numbers aren't wrong, you're just not reading the chart right.

The chart doesn’t even show all the teams. It cherry picks the ones that re enforce their argument. Weak data.

 

[nexttothemoon]

You do realize that the GF% vs Points per game (2016/17) chart would have a R squared of:

0.948741

[TBODY] [/TBODY]

Here's all the actual R squared stats from last season in the NHL:

R-squared is a statistical measure of how close the data are to the fitted regression line. It is also known as the coefficient of determination, or the coefficient of multiple determination for multiple regression. ... 100% indicates that the model explains all the variability of the response data around its mean.

CF/60

CA/60

CF%

FF/60

FA/60

FF%

SF/60

SA/60

SF%

GF/60

GA/60

GF%

SCF/60

SCA/60

SCF%

SCGF/60

SCGA/60

SCGF%

SCSH%

SCSV%

HDCF/60

HDCA/60

HDCF%

HDGF/60

HDGA/60

HDGF%

HDSH%

HDSV%

SH%

SV%

PDO

[TBODY] [/TBODY]

0.1773

0.0093

0.118

0.159

0.1041

0.195

0.261

0.125

0.282

0.7238

0.672

0.948

0.3762

0.00746

0.283

0.577972

0.62597

0.88799

0.32780

0.52157

0.266628

0.001182

0.233705

0.407168

0.446546

0.645152

0.174067

0.346533

0.63

0.50

0.788

[TBODY] [/TBODY]

So in order these were the "best" correlative stats last season with regards to pts/game:

GF% .949
Scoring Chances GF% .888
PDO .789
GF/60 .724
GA/60 .673
High Danger GF% .645
SH% .639
Scoring Chances GA/60 .626
Scoring Chances GF/60 .578
Scoring Chances SV % .522
SV% .506
High Danger GA/60 .447
High Danger GF/60 .407
Scoring Chances F/60 .376
High Danger SV% .347
Scoring Chances SH% .328
Scoring Chances% .284
Shots F% .283
High Danger Chances F/60 .267
Shots F/60 .262
High Danger Chances F% .234
Fenwick F% .196
Corsi F/60 .177
High Danger SH% .174
Fenwick F/60 .160
Shots A/60 .125
Corsi F% .118
Fenwick A/60 .104
Corsi A/60 .009
Scoring Chances A/60 .007
High Danger Chances A/60 .001



So essentially it's obvious to see that stats based around reality (goals for and against on the scoreboard) are the most indicative of a teams success.

The ability to actually score goals and keep them out is very highly correlated to pts... common sense really.

Corsi stats have very low correlation to actual scoreboard results... which should be easily deduced from the number of times a team can out-corsi another team yet still loses the game... ie the Oilers this season for a prime example right under our noses.

 

[Little Fury]

So essentially it's obvious to see that stats based around reality (goals for and against on the scoreboard) are the most indicative of a teams success.

The ability to actually score goals and keep them out is very highly correlated to pts... common sense really.

You realize we're not talking about the same thing here, right? The issue isn't which measure is correlated with success, but which is most predictive of future success.